An early-onset periodontitis cohort was used to investigate the degree of bias in estimates of prevalence and severity of periodontal attachment loss produced due to various partial recording protocols (PRP's). 266 subjects, ages 19-25 years, were studied. A complete examination consisted of a maximum of 168 measurements, six sites per tooth on all 28 teeth. However, PRP's are routinely used in national surveys of oral health because of logistical considerations. Thus, there is substantial interest in obtaining estimates of the true prevalence and severity of periodontal diseases based on the NIDCR PRP used in our national surveys. The PRP's considered in this study included full mouth assessments for the following combinations of sites: MB, B, DB, ML, L, DL, separately, (MB, B), (MB, B, DB), (MB, B, DL), (ML, L, DL) and (MB, DB, BL, DL). Prevalence of disease was systematically underestimated by all PRP's. The prevalence of attachment loss greater than 3 mm was 64.7 percent based on the full mouth score. However, the corresponding estimates were: 20 to 25 percent for mid-tooth sites, 43.6 to 45.1 percent among the 4 interproximal sites, individu-ally; 47.7 percent for the (MB, B) pair, 54.5 to 56.8 percent for the triple site combinations, and 62 percent for the interproximal quadruplet. Attachment loss severity was 1.17 mm based on the full mouth score. The corresponding estimates were: 0.62 and 0.66 mm for mid-tooth sites, 1.39 to 1.48 mm among the 4 interproximal sites, individually; 1.07 mm for the (MB, B) pair, 1.15 to 1.18 mm for the triple site combinations, and 1.43 mm for the interproximal quadruplet. For early onset periodontitis subjects the prevalence of attachment loss (greater than 3 mm) was underestimated by roughly 15 percent for triple site scores, 30 percent for the (MB, B) and interproximal single site scores, and over 60 percent for the single mid-tooth scores. For severity of disease there was a slight bias (less than 2 percent) for the triple site scores, a 9 percent underestimate for the (MB, B) pair, a 20 to 25 percent overestimate by interproximal single site scores, and a 45 percent underestimate for the single mid-tooth site scores. These studies of the effects of partial recording help elucidate the real patterns of periodontal disease estimates obtained in the NHANESIII survey and also assist in the selection of a PRP for use in the NHANESIV survey.